Recommendations
1 to acquire · 0 to dispose
| Ticker | Name | Close | Action | Action Price Where to act — a buy-limit on a pullback or a buy-stop above resistance, not the current price. Sanity-check vs Close: on a held name, an action price stranded above the close means the queued level was set when the stock was higher — stale, awaiting reprice. | Stop The risk-defined exit, set below the entry for a long (~1.2–1.8×ATR). A held long whose Stop sits at/above the Close is already below its stop — a stale box awaiting reprice. | 1wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | 4wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | 12wk Model estimate of the likely price at this horizon — a forward projection, not a guarantee, and separate from the trade's Action Price or Stop. Where the ML estimator is enabled these are per-ticker calibrated forecasts (~80% coverage band) that are refreshed periodically and can drift between refreshes; otherwise they're interpolated from the signal's swing target. Confirm against the settled close. | Rationale |
|---|---|---|---|---|---|---|---|---|---|
| 1810 | XIAOMI CORP-W | 28.46 | BUY | 28.46 | 27.80 | 29.19 (+2.6%) | 30.90 (+8.6%) | 28.50 (+0.1%) | Open a new 4,000-share position in Xiaomi at HK - the book's first trade in eighteen sessions. Xiaomi printed BUY at conviction 69 and rank 1 with the entry sitting exactly at the market, so we are paying the model's price rather than chasing a move: the shares closed only 2.4% above VWAP with RSI at 56, mid-range, on volume 75% above the prior session. This is the same name the book refused on 21 August at HK, when it was a chase of a post-earnings pop; that refusal saved 4.3%, and the lesson it invoked said to wait for the pullback to the model entry. The pullback has now happened and we are buying 1.9% cheaper, with the twelve-week view flipped from negative to positive. The business case is mixed but not broken: Q2 revenue rose 30.5% to RMB108.9bn with record smartphone pricing and a sixth straight quarter of EV delivery growth, against a genuine margin squeeze - handset gross margin fell from 11.5% to 8.5% as AI demand pushed memory prices up. Crucially, Xiaomi is the only actionable name in this book that PASSES our accounting-quality screen, where every candidate we have turned down for six weeks failed it outright. The trade is underwritten by the four-week target of 30.90 (+8.6%) and a reward-to-risk near 3.7, not by the twelve-week view, which at +0.1% is honestly flat. First resistance is EMA100 at 28.87, 1.4% overhead. OPERATIONAL: adj cv 68 = raw 69 + val_adj -1 (GROWTH_PRICED_IN, MoS -0.19097 on the 08-26 refreshed CSV) + scorecard_bias 0 (HT-2; FailedBreakdown_Up_20D|buy n=2 report-only, hit_4w 0.00 - the L1 handicap is stated, not ignored). Quality_Pass TRUE (F-Score 6) - gate PASSES, the one thing separating this from 6869 (F6/no DCF/12.5x EPV) and 1211 (F4), both hard-blocked. HT-4: T12 28.50 >= Close 28.46 by 4 cents = MARGINAL pass; T4 30.90 = +8.57%; R:R 3.70. Entry_Distance_ATR 0.00, Order_Type Market, entry range 28.16-28.77 (+/-0.25 ATR). Sizing: 28.39% x HKx 20% = HKsize_mod 1.00 (MoS -0.191 > -1), /28.46 = 4,051 -> lot 200 -> = HK = 5.61% of book; max_position_pct 60% and both non-binding. Risk to the 27.80 stop = HK = 0.130% of book. WEAKNESS RECORDED: stop 27.80 is 0.54 ATR and is EXACTLY today's low (Systemic #3/#5) - sub-noise and likely to be chopped; the mitigation is the 0.13% loss size, one-tenth of what 6869's 3.39-ATR stop would have risked. Structure: below EMA100 28.87 (-1.4%) and EMA200 32.99 (-13.7%), above a rising EMA50 27.33 (+4.1%) - early base reclaim, not an established uptrend. Close sits 82% up the Est band 25.35-29.15. Label trail BUY(78)->WATCH(45)->BUY(69) but the signal FAMILY and the Consolidation_Bullish consensus were continuous across all three readings; Buy_Rank 1 on both BUY prints. COUNTERFACTUAL APPLIED: no_chase_queue n=6 median +9.01%/1wk [cost] is driven by QUEUED PULLBACKS, the class 1810 belongs to today - not the post-parabolic chases this book correctly refuses (per the 08-25 entry). This is the trade that distinction was drawn to permit. EXECUTION GUARD (L8, numeric): fills at the 2026-08-27 open ONLY if the open is inside 28.03-28.89. Open < 28.03 (-1.5%) -> STAND DOWN, do not buy, re-read at the 08-27 EOD (NVIDIA reports tonight, HK reacts 08-27; 1810 is NOT on the horizon read-through list but a 2% adverse HSTECH open puts the 27.80 stop through at the bell). Open > 28.89 (+1.5%) -> STAND DOWN, no chase. f=0.30 per L5 (calib_1w +0.129 unchanged on a stale n=10 sample); 1wk 29.19 = 28.46 + 0.30 x 2.44. 12wk 28.50 carries the DCF annotation: exceeds DCF anchor 17.52 - projection assumes growth beyond priced-in. Yahoo 1810.HK/news/ returned HTTP 503 on direct fetch (recurring); the news read rests on search-index coverage. · news |
| 0992 | Lenovo Group Limited | 30.56 | HOLD | 30.56 | 29.40 | 31.59 (+3.4%) | 34.00 (+11.3%) | 43.55 (+42.5%) | Hold the full of Lenovo and leave the trailing stop where it is. The shares eased 0.8% to 30.56 on volume 62% lighter than yesterday - a quiet give-back of a third of the prior session's gain, not distribution - and the unrealised gain stands at +11.1%, or HKThe model's conviction fell sharply, from, and that is recorded rather than glossed; but the trend pillar has now held at its maximum reading for a seventh consecutive session, the bullish consensus and trending regime both held, the twelve-week view is +42.5%, and the shares still trade 33% below their cash-flow anchor of 45.62 with a clean quality screen. The 13 August results remain the thesis: an AI-server pipeline of US$54bn, up 157% quarter-on-quarter, and infrastructure revenue close to doubled. Netlist's ITC patent complaint is twelve days old, already in a rising tape and measured in quarters, not sessions. Nothing in the next ten trading days is scheduled against the name. There is no exit case on any test we run. OPERATIONAL: TRAIL HELD AT 29.40 FOR A FOURTH SESSION. NOT raised to the row's printed 30.04 (0.30 ATR under the market - the same self-triggering defect rejected at 0.27 ATR yesterday). The chandelier is DEFECTIVE IN THE INVERTED DIRECTION AGAIN: ATR 1.9670 -> 1.7176, so 36.16 - 3.0 x 1.7176 = 31.007, ABOVE the 30.56 close = unexecutable (Systemic #1, fourth consecutive session). Cross-check: a principled close - 1.5 x ATR clamp gives 27.98, so the carried 29.40 is ALREADY TIGHTER and must not be tightened further. Headroom 3.80%, protecting HKat 6.92% above the 27.498 cost. PROFIT-TAKE (ADR-0012): NOT ARMED - Exit_Warning blank, Gain_ATR 1.78 < 3.0, T12_Progress_Pct 70.17 < 90 AND FROZEN FOR A TENTH SNAPSHOT (hand-computed 19.1%). The -22 pt conviction drop IS a textbook fire tick (b) and it CANNOT FIRE because arming failed on a known defect - Systemic #2 demonstrably suppressed a live ADR-0012 signal today. L2: close 30.56 ABOVE the 29.40 trail and 12wk +42.50% positive - both legs fail, no L2 exit. ADR-0020: live stop 29.40 = peak stop 29.40, 0% erosion, no cut, not triggered (hk carries no trail_erosion config). G1/G3 would block any valuation/divergence trim regardless (Trend 77, Regime Trending, 12wk +42.5%). val_adj -3 (MoS -0.6517 on the refreshed CSV) is deep, recorded and NON-ORIGINATING per CLAUDE.md section 9. T12 43.55 < DCF 45.62 so NO DCF annotation. NO ADD - the re-arm needs a bullish ACTIONABLE label; HOLD is not actionable and conviction fell 22 points. BREACH RULE: a decisive close below 29.40 is the exit. Est band 29.95-31.17, close at 50% of band. f=0.30 per L5. · news |
| 6869 | YOFC | 163.00 | WATCH | 167.33 | 122.32 | n/a | 257.35 (+53.8%) | 114.34 (-31.7%) | Declined for a second day, and the tape has already begun pricing yesterday's refusal. YOFC opened at 164.60, ran to 174.50 - up 7.4% - and closed at 163.00, in the bottom 6% of its own range: a complete intraday round-trip of the whole advance, one session after an unexplained 18% blow-off. The model's conviction collapsed from in a single reading. The valuation refresh made the case worse, not better: on today's price the shares sit further beyond any defensible anchor than they did yesterday, and the company still fails our accounting-quality screen outright with no reliable cash-flow anchor, changing hands at roughly twelve and a half times its earnings-power value. The model's own twelve-week view is 114.34, thirty per cent BELOW today's close, while the four-week target on the same row says 257.35 - an eighty-eight point contradiction that is a modelling defect, not a forecast. The risk line remains unusable: a stop at 122.32 is 27% below the buy-stop entry, and NVIDIA's results tonight plus YOFC's own 31 August interim deadline are two binary events inside four sessions that such a stop can only sit through. Re-arm on a pullback into 140-150 that holds EMA100 on the retest, AFTER the 31 August interim, with a positive twelve-week view and a stop inside 1.5 ATR - never at today's price. OPERATIONAL: adj cv 53 = raw 63 + val_adj -10 (MoS -11.4728 on the 08-26 refreshed CSV, WORSE than the -9.4876 recorded yesterday exactly as predicted) + bias 0 = TWELVE POINTS SUB-FLOOR - unlike yesterday, the fusion floor now kills it on its own. Quality_Pass FALSE (F-Score 6, no DCF, EPV 13.07) = hard block on a -0 new entry. HT-4 forward-view leg fails (T12 -29.85%). Order_Type is now correctly a Buy Stop @167.33 ABOVE the 163.00 close (Entry_Distance_ATR +0.30) - yesterday's below-market buy-stop artefact is gone - but stop 122.32 = 45.01 from entry = 3.08 ATR = -26.9%. VWAP_Distance +28.13%, RSI 60.71, still 46.6% below the 52W high 305.00 and only +0.77% above EMA100 161.75. Counterfactual sizing for the grader: 28.39% x HKx 20% = HKx 0.5 (MoS <= -3) = HK -> /163.00 = -> 0 at a. Not. Horizon: on the NVIDIA read-through list (08-26 US close, HK reaction 08-27, band +/-12%) AND named in the 08-31 HK interim-results deadline cohort (band +/-6%). · news |
| 1211 | BYD | 92.60 | WATCH | — | — | n/a | n/a | 94.93 | Off the table - the buy label is gone. BYD printed BUY at conviction 69 for two readings running and today reverted to NO TRADE at conviction zero, with the engine citing consensus conflict and low conviction. The bullish consensus flipped to conflicted and the momentum pillar halved. Yesterday's re-arm condition - a third consecutive actionable BUY after the 31 August interim - is therefore moot, because there was no third print. Interim results are dated 29 August and first trade on 31 August; we would rather see the print in any case. OPERATIONAL: Action BUY cv69 -> NO TRADE cv0; NoTrade_Reason verbatim: 'Consensus conflict + low conviction'. Pillar_Consensus Consolidation_Bullish -> Conflicted; Momentum; Entry/Stop/Target all blank. Quality_Pass still FALSE (F-Score 4, no DCF, EPV 38.15); refreshed MoS -1.4270 (val_adj would be -7). Close 92.60 still below EMA200 96.46. No re-arm carried - the name must produce a fresh actionable label after 08-31. · news |
| PORTFOLIO | hktech | — | WATCH | — | — | n/a | n/a | n/a | ONE ACQUISITION, NO DISPOSALS - the seventeen-session cash freeze ends. We are buying of Xiaomi at HK, HKor 5.6% of the book, funded entirely from. The reason this trade clears where six weeks of candidates did not is simple: Xiaomi is the only actionable name in this book that passes our accounting-quality screen, and it is being bought at the model's own entry price after a pullback - not chased. The book refused this same name at HKon 21 August as a post-earnings chase, saved 4.3% doing so, and is now buying it 1.9% cheaper with the twelve-week view turned positive. Risk is trivially small - HKor 0.13% of the book, to the stop - and the trade is underwritten by the four-week target, not by a twelve-week thesis, which at +0.1% is flat. YOFC was declined for a second day and immediately round-tripped its entire intraday gain, closing at the bottom of its range after touching +7.4%; BYD's buy label disappeared altogether. On the sell side there is nothing to sell: Lenovo's conviction fell but its trend, consensus, regime, twelve-week view and clear ten-session event calendar are all intact, and its trailing stop stays at 29.40 for a fourth session. OPERATIONAL: 1 ACQUIRE / 0 DISPOSE. Book HK ( 15.05%, cash 84.95%), +1.53% vs the 2,000,000 seed. HK -> HKafter the fill (6.6%); post-execution HK = 20.7%, cash 79.3%. PENDING for the 2026-08-27 open with a numeric gap guard: fill 4,000 x 1810 at market ONLY if the open is inside 28.03-28.89; stand down outside it (NVIDIA reports tonight, HK reacts 08-27). No GTC queue open anywhere; nothing to reprice, withdraw or expire. Anchor is a settled 17:00 HKT close (L4 authoritative) and the 08-20/08-24 coverage gap did NOT recur - today's scan ran on schedule; those two sessions remain permanently un-back-writable. VALUATION REGENERATED: prior FetchedAt 2026-08-19 was 7.00 days, over the bar; new FetchedAt 2026-08-26 17:19:02, marked to today's closes. Scorecard: entries 119, graded 10, pending 0, unfilled 1, calib_1w +0.129; NO actionable cells, all biases 0 (HT-2); f=0.30 retained per L5. Overlay WEEK 2026-07-05 is 52 days old, badly stale and inert; the 0700 override is moot at. Learnings ACTIVE block is 39 days old and FIVE PROPOSED blocks (07-25, 08-01, 08-08, 08-15, 08-22) remain inert and were NOT applied; Step 8a skipped (batch mode). |
Outcome & track record
Accuracy at the time of this record.
10 Graded decisions Past recommendations old enough to score against what actually happened (target hit / stopped out / closed). This is the evidence base — the bigger the number, the more history behind the calls. Click to inspect every one. See the decisions →
0.129 1-week calibration How much of the eventual 4-week move typically shows up in the first week (0 = almost none, 1 = most of it). Use it to read the 1-week estimates: a low number means prices drift toward the target slowly, so don't over-read a soft week-1.
50 Published records How many dated end-of-day verdicts exist for this book — the length of the public track record you can browse in the Archive below.